An application method, system, device and medium based on the combination of AI and IoT scenarios, cloud computing and big data

Abstract

The application discloses a kind of based on AI and the combination application method, system, equipment and medium of cloud computing and big data under the combination scene of IoT, belong to intelligent monitoring technical field, include: the observation area is divided into center area and edge area;Based on historical data, the corresponding relationship of incomplete mode of feature in edge area and auxiliary detection information is established;Obtain the observation information of current target;In response to current target being located in edge area, identify incomplete mode of feature, query auxiliary detection information, generate compensation observation information according to narrowband internet of things sensor trigger information compensation visual feature loss;Determine whether to meet the same target condition;Generate target trajectory record and update feature compensation associated data.The application compensates AI camera edge area visual feature loss using narrowband internet of things sensor trigger information, and realizes cross-camera accurate tracking based on cloud computing and big data.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, specifically to a method, system, device, and medium for the combined application of cloud computing and big data in a scenario combining AI and IoT. Background Technology

[0002] In intelligent monitoring systems, AI cameras capture images of targets at the edge of the field of view, resulting in distortion and reduced resolution. Extracted visual features suffer from dimensional loss, making it difficult to accurately correlate targets with subsequent camera observations once they leave the edge area. While narrowband IoT sensors can detect target-triggered events, they cannot provide identification information. Existing methods typically use sensor triggering only as auxiliary verification, failing to fully exploit the compensating value of sensor information for visual features, leading to a high failure rate in cross-camera tracking of targets at the edge. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention provides a method, system, device and medium for the combined application of cloud computing and big data in the scenario of AI and IoT integration.

[0004] Therefore, the technical problems solved by this invention are: how to solve the problem of high failure rate of cross-camera target tracking due to the lack of visual feature dimensions in the edge area of ​​AI cameras, and the problem that the triggering information of narrowband IoT sensors is not used to compensate for the lack of visual features.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for the combined application of cloud computing and big data in a scenario combining AI and IoT, comprising, The observation area is divided into a central area and an edge area, which is the field of view of the AI ​​camera for the target monitoring scene; Establish feature compensation association data, which includes a first mapping relationship, the first mapping relationship being used to map the feature incompleteness pattern of the edge region and auxiliary detection information; In response to the target being detected being located in an edge region, compensated observation information of the target being detected is obtained. The process of obtaining the compensated observation information of the target being detected includes: querying the feature incomplete pattern of the first observation information of the target being detected according to the first mapping relationship, obtaining auxiliary detection information mapped by the feature incomplete pattern, and generating the compensated observation information based on the auxiliary detection information. In response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition, a target trajectory record is generated, and the feature compensation association data is updated based on the target trajectory record; The first observation information and the second observation information are acquired through a visual acquisition device, and the auxiliary detection information is acquired through an Internet of Things (IoT) sensor.

[0006] As a preferred embodiment of the method for combining cloud computing and big data in an AI and IoT scenario described in this invention, wherein generating the compensated observation information based on the auxiliary detection information includes: Obtain the actual auxiliary detection information between the first observation information and the second observation information; Based on the second mapping relationship, and combined with the actual auxiliary detection information, query the feature differences mapped by the actual detection information; The first observation information and the feature difference are fused to generate the compensated observation information. The effective features of the first observation information in the compensated observation information retain their original values, and the missing features of the first observation information are filled by the feature difference.

[0007] As a preferred embodiment of the cloud computing and big data integration application method based on AI and IoT scenarios described in this invention, the division of the edge region further includes: The edge region is divided into multiple edge sub-regions along the azimuth angle based on the observation parameters; The incomplete feature pattern includes edge sub-region identifiers and feature missing information.

[0008] As a preferred embodiment of the method for combining cloud computing and big data in an AI and IoT scenario described in this invention, the determination of the same target condition includes: When both the first observation information and the second observation information are complete feature information, determine whether the first observation information and the second observation information satisfy the same target condition; When the first observation information is incomplete feature information and the second observation information is complete feature information, determine whether the compensation observation information and the second observation information satisfy the same target condition. When the first observation information is complete feature information and the second observation information is incomplete feature information, feature compensation is performed on the second observation information to generate second compensated observation information, and it is determined whether the first observation information and the second compensated observation information satisfy the same target condition. When both the first observation information and the second observation information are incomplete feature information, it is determined whether the common valid features of the compensated observation information and the second compensated observation information satisfy the same target condition.

[0009] As a preferred embodiment of the cloud computing and big data combined application method in an AI and IoT scenario described in this invention, after updating the feature compensation correlation data based on the target trajectory record, the method further includes: Extract endpoint observation information from the target trajectory record; Query multiple historical auxiliary detection information in the updated feature compensation association data; Extract the first detection unit identifier from each of the aforementioned historical auxiliary detection information; The frequency of occurrence of each of the first detection unit identifiers is counted, and the first detection unit identifier with the highest frequency of occurrence is selected as the predicted detection unit identifier; Generate and output a prediction record that includes the prediction detection unit identifier and the endpoint observation information.

[0010] Generate and output a prediction record that includes the prediction detection unit identifier and the endpoint observation information.

[0011] As a preferred embodiment of the method for combining cloud computing and big data in an AI and IoT scenario described in this invention, the step of establishing feature-compensated correlation data includes: Extract complete historical feature information, incomplete historical feature information, and historical auxiliary detection information from historical data; Calculate the feature difference between the complete historical feature information and the incomplete historical feature information, and establish a second mapping relationship between the historical auxiliary detection information and the feature difference; Extract the incomplete feature patterns from the historical incomplete feature information, and establish the first mapping relationship between the incomplete feature patterns and the historical auxiliary detection information; The feature compensation association data includes the first mapping relationship and the second mapping relationship.

[0012] As a preferred embodiment of the combined application method of cloud computing and big data in the AI ​​and IoT scenario described in this invention, the first observation information is the visual feature information of the target to be detected collected by the AI ​​camera at the first observation time; the second observation information is the visual feature information of the target to be detected collected by the AI ​​camera at the second observation time; the auxiliary detection information is the detection information reported by the IoT sensor between the first observation time and the second observation time; and the observation parameters are the field of view parameters of the visual acquisition device.

[0013] This invention provides a combined application system of cloud computing and big data based on the combination of AI and IoT scenarios.

[0014] To address the aforementioned technical problems, the present invention provides the following technical solution: a cloud computing and big data combined application system based on AI and IoT scenarios, comprising: a visual acquisition device, an IoT sensor, and a cloud server; The cloud server includes: The region division module is used to divide the observation area into a central region and an edge region, wherein the observation area is the field of view of the AI ​​camera for the target monitoring scene; The association establishment module is used to establish feature compensation association data, which includes a first mapping relationship. The first mapping relationship is used to map the feature incomplete pattern of the edge region and auxiliary detection information. The feature compensation module is used to respond to the fact that the target to be detected is located in the edge region, query the feature incomplete pattern of the first observation information of the target to be detected according to the first mapping relationship, obtain the auxiliary detection information mapped by the feature incomplete pattern, and generate compensated observation information based on the auxiliary detection information. The determination module is used to generate a target trajectory record in response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition; The update module is used to update the feature compensation association data based on the target trajectory record.

[0015] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for combining cloud computing and big data in a scenario combining AI and IoT.

[0016] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for the combined application of cloud computing and big data in a scenario combining AI and IoT.

[0017] The beneficial effects of this invention are as follows: By using sensor-triggered sequences to compensate for the missing visual feature dimensions in the edge regions of the camera, observations of edge regions that were originally unrelated due to feature incompleteness can also participate in target tracking, thus expanding the effective monitoring range of the camera. Compared to existing methods that only use the complete features of the central region of the field of view, this invention makes the observation information of the edge regions of the field of view usable, avoiding tracking loss caused by edge observation failure when the camera switches.

[0018] This invention makes full use of the narrowband IoT sensors already deployed in the monitoring scenario, transforming sensor trigger information from simple event recording into a source of visual feature compensation, thereby improving cross-camera tracking capabilities without increasing equipment investment.

[0019] This invention predicts the location of sensors that may be triggered by the target next based on updated correlation data, enabling the monitoring system to change from a passive response to an active early warning, thus gaining a time window for security personnel to handle the situation on-site or for the automatic control system to coordinate operations. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the overall application method of cloud computing and big data in a scenario combining AI and IoT, as provided in one embodiment of the present invention.

[0022] Figure 2 This is a flowchart illustrating the establishment of a general process feature compensation relationship for a method for the combined application of cloud computing and big data in an AI and IoT scenario, as provided in one embodiment of the present invention.

[0023] Figure 3 The flowchart below shows a second mapping relationship establishment method for a cloud computing and big data application method based on the combination of AI and IoT in an embodiment of the present invention. Detailed Implementation

[0024] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides a method for the combined application of cloud computing and big data in a scenario combining AI and IoT, including: S1: Divide the observation area into a central area and an edge area, whereby the observation area is the field of view of the AI ​​camera for the target monitoring scene.

[0026] In some embodiments, step S1 includes: S11, the observation area is divided into a central area and an edge area according to the observation parameters, and the edge area is divided into multiple edge detection sectors along the azimuth angle.

[0027] For example, step S11 includes steps S111 to S114, wherein: S111, determine the spatial range of the central region and the spatial range of the edge region based on the observation parameters.

[0028] In some embodiments, the observation area is the field of view of the visual acquisition device, and the observation parameters are field of view parameters. For example, the visual acquisition device is an AI camera, and the field of view parameters include the field of view angle and the effective detection distance.

[0029] S112, based on the statistical analysis of historical monitoring data on the cloud computing platform, determines the preset distance threshold.

[0030] For example, step S112 includes steps S1121 to S1125, wherein: S1121, The cloud server extracts multiple target movement records from the historical monitoring data.

[0031] S1122, Calculate the proportion of effective feature dimensions to the total number of dimensions in the visual feature vector at each radial distance position. It should be noted that the visual feature vector includes multiple feature dimensions. When the target to be detected is located at the edge of the field of view, due to the influence of lens edge imaging quality, the values ​​of some feature dimensions are zero.

[0032] S1123, when the proportion of the number of effective feature dimensions to the total number of dimensions is lower than a preset proportion threshold, the radial distance position is marked as a feature missing position. Optionally, the preset proportion threshold is 85% to 95%, preferably 90% in AI and IoT combined monitoring scenarios.

[0033] S1124, the cloud server filters target movement records marked as feature missing locations, extracts the corresponding data of narrowband IoT sensor trigger information and visual feature missing dimensions in the target movement records, and uses the correlation analysis algorithm of the cloud computing platform to determine whether the narrowband IoT sensor trigger information can provide compensation data for the visual feature missing dimensions.

[0034] S1125, the minimum radial distance position is extracted from the target movement record that can provide compensation data as the preset distance threshold. It should be noted that, through statistical analysis of historical monitoring big data on a cloud computing platform, when the preset ratio threshold is 90%, the ratio between the preset distance threshold and the effective detection distance is 68% to 72%.

[0035] S113, the central region is the region within the field of view whose radial distance from the optical axis of the AI ​​camera is less than the preset distance threshold, and the edge region is the region within the field of view whose radial distance is greater than or equal to the preset distance threshold.

[0036] S114, the edge region is uniformly divided into multiple edge detection sectors along the azimuth angle, and a sector identifier is assigned to each edge detection sector. In some embodiments, the cloud server queries the total number of narrowband IoT sensors from the sensor deployment database and sets the number of edge detection sectors to the square root of the total number of narrowband IoT sensors, rounded up. Optionally, the number of edge detection sectors is 4 to 12, preferably 8 in AI and IoT combined monitoring scenarios.

[0037] Through the implementation of step S1, the observation area is divided into a central area and an edge area based on the statistical analysis of historical monitoring big data on the cloud computing platform. The spatial boundary where the visual feature dimension begins to be missing is determined, so that subsequent steps can establish a compensation mechanism for the feature missing problem in the edge area. This provides a spatial division basis for solving the problem of high failure rate of cross-camera target tracking caused by the lack of visual feature dimension in the edge area of ​​AI camera.

[0038] S2: Establish feature compensation association data, which includes a first mapping relationship. The first mapping relationship is used to map the feature incomplete pattern of the edge region and auxiliary detection information.

[0039] In some embodiments, step S2 includes steps S21 to S24, wherein: S21, extract complete historical feature information, incomplete historical feature information, and historical auxiliary detection information from historical data. It should be noted that the historical data is stored in a big data storage system on a cloud server.

[0040] For example, step S21 includes steps S211 to S213, wherein: S211, the cloud server extracts the target movement record from the historical data. In some embodiments, the target movement record includes a historical center region feature vector, a historical edge incomplete feature vector, a historical starting sector identifier, a historical ending feature vector, and a historical sensor trigger sequence.

[0041] S212, historical starting point observation data and historical ending point observation data are extracted from the target movement record. The cloud server determines whether the target to be detected is located in the central region or the edge region in the historical starting point observation data. Further, when the target to be detected is located in the central region in the historical starting point observation data, the visual feature vector extracted from the historical starting point observation data is used as historical complete feature information, which is the historical central region feature vector. Even further, when the target to be detected is located in the edge region in the historical starting point observation data, the visual feature vector extracted from the historical starting point observation data is used as historical incomplete feature information, which is the historical edge incomplete feature vector.

[0042] S213, extract IoT sensor trigger records between the start observation time and the end observation time from the target movement record, and extract the narrowband IoT sensor identifier and trigger time from the IoT sensor trigger records to form historical auxiliary detection information. In some embodiments, the auxiliary detection information is obtained through an IoT sensor, the IoT sensor being a narrowband IoT sensor, and the historical auxiliary detection information includes a historical sensor trigger sequence.

[0043] S22, calculate the feature difference between the historical complete feature information and the historical incomplete feature information, and establish a second mapping relationship between the historical auxiliary detection information and the feature difference.

[0044] For example, step S22 includes steps S221 to S223, wherein: S221, the cloud server filters records from the target movement records that simultaneously include both complete historical feature information and incomplete historical feature information.

[0045] S222, calculate the feature dimension difference vector between the complete historical feature information and the incomplete historical feature information.

[0046] Furthermore, the value of each dimension of the feature dimension difference vector is the value of the historical complete feature information in that dimension minus the value of the historical incomplete feature information in that dimension. Even further, when the value of the historical incomplete feature information in a certain feature dimension is zero, the value of that feature dimension in the feature dimension difference vector is equal to the value of the historical complete feature information in that dimension. In some embodiments, the feature dimension difference vector represents the feature missing pattern caused by edge detection sectors.

[0047] S223, extract historical auxiliary detection information from the target movement record, and establish a second mapping relationship between the historical auxiliary detection information and the feature dimension difference vector.

[0048] In some embodiments, narrowband IoT sensor identifiers are extracted from the historical sensor trigger sequence to form a historical sensor identifier sequence, and a mapping relationship between the historical sensor identifier sequence and the feature dimension difference vector is established and stored in a sequence compensation mapping table.

[0049] S23, extract the feature incomplete pattern of the historical incomplete feature information, and establish a first mapping relationship between the feature incomplete pattern and the historical auxiliary detection information.

[0050] For example, step S23 includes steps S231 to S233, wherein: S231, identify the feature dimension numbers with zero values ​​in the incomplete historical feature information to form a set of historical missing dimension numbers.

[0051] S232, extract the edge detection sector identifiers from the sources of the historical incomplete feature information, and concatenate the edge detection sector identifiers with the historical missing dimension number set to form a feature incomplete pattern. In some embodiments, the feature incomplete pattern includes edge detection sector identifiers and feature missing information, where the feature missing information is the historical missing dimension number set. For example, the historical starting point sector identifier is concatenated with the historical missing dimension number set to form a historical edge missing identifier.

[0052] S233, extract historical auxiliary detection information from the target movement record from which the historical incomplete feature information originates, and establish a first mapping relationship between the incomplete feature pattern and the historical auxiliary detection information. In some embodiments, establish a mapping relationship between the historical edge missing identifier and the historical sensor identifier sequence and store it in an edge sequence mapping table.

[0053] S24, the feature compensation association data includes the first mapping relationship and the second mapping relationship.

[0054] Through the implementation of step S2, the trigger information of narrowband IoT sensors is transformed from simple event recording into a source of visual feature compensation. A mapping relationship is established between the sensor trigger sequence and the missing dimensions of visual features, enabling the sensor trigger data, which originally could not provide identification information, to have the ability to compensate for the missing dimensions of visual features. The feature compensation value of deployed narrowband IoT sensors is explored without increasing equipment investment.

[0055] S3: In response to the target being detected being located in an edge region, compensated observation information of the target being detected is obtained. Obtaining the compensated observation information of the target being detected includes: querying the feature incomplete pattern of the first observation information of the target being detected according to the first mapping relationship, obtaining auxiliary detection information mapped by the feature incomplete pattern, and generating the compensated observation information based on the auxiliary detection information.

[0056] In some embodiments, step S3 includes two parts: acquiring first observation information and second observation information of the target to be detected, and generating compensated observation information, specifically including steps S31 to S39, wherein: S31, the AI ​​camera captures image frames and uploads them to the cloud server.

[0057] S32, the cloud server extracts AI observation records from the image frame, including the acquisition feature vector, AI camera identifier, and observation time.

[0058] S33, combining the spatial ranges of the central and edge regions determined in step S1, it is determined whether the target to be detected is located in the central or edge region. Further, the radial distance between the target to be detected and the optical axis of the AI ​​camera is calculated based on the pixel coordinates of the target in the image frame. Even further, when the radial distance is less than the preset distance threshold, it is determined that the target to be detected is located in the central region, the acquired feature vector in the AI ​​observation record is marked as the central region feature vector, and the AI ​​observation record is marked as a center type.

[0059] S34, when the target to be detected is located in the edge region, the feature dimensions of the collected feature vector are classified into a set of attenuated dimension numbers and a set of retained dimension numbers by the edge attenuation algorithm.

[0060] For example, step S34 includes steps S341 to S345, wherein: S341, the cloud server extracts the sector identifier of the edge detection sector where the target to be detected is located, and filters all mapping relationships that match the sector identifier in the edge sequence mapping table to obtain N candidate historical edge missing identifiers and their corresponding N historical sensor identifier sequences. It should be noted that the edge sequence mapping table stores the first mapping relationship between historical edge missing identifiers and historical sensor identifier sequences established in step S2. Each historical edge missing identifier includes a historical starting sector identifier and a set of historical missing dimension numbers.

[0061] S342, query the sensor deployment database for the expected trigger sensor set for the target location. It should be noted that the narrowband IoT sensors are deployed at fixed spatial locations within the monitoring area, and the sensor deployment database stores the identifier, deployment coordinates, and detection range parameters of each narrowband IoT sensor. Further, the cloud server calculates the spatial coordinates of the target based on its radial distance and azimuth angle, extracts the spatial coordinate sequence of the target in the most recent K frames of images from the real-time observation data cache (K is preferably 3 to 5), calculates the ratio of the distance difference to the time difference between adjacent spatial coordinates to obtain the instantaneous moving speed of the target, determines the target's reachable distance based on the instantaneous moving speed and the observation time interval, determines the reachable spatial range with the target location as the center and the reachable distance as the radius, queries the sensor deployment database for the narrowband IoT sensors within the reachable spatial range, and extracts the identifiers of these sensors to form the expected trigger sensor set.

[0062] S343, calculate the spatial matching degree between each historical sensor identifier sequence and the expected triggering sensor set. Further, calculate the proportion of the number of sensor identifiers appearing in the expected triggering sensor set relative to the total length of the historical sensor identifier sequence, as the spatial matching degree of that historical sensor identifier sequence.

[0063] S344, Select the historical sensor identifier sequence with the highest spatial matching degree, and extract the set of historical missing dimension numbers from the historical edge missing identifier corresponding to the historical sensor identifier sequence, as the current attenuation dimension number set. Further, after removing the attenuation dimension number set from all feature dimension numbers of the acquired feature vector, the remaining feature dimension numbers constitute the retained dimension number set.

[0064] S345, set the feature dimension values ​​belonging to the attenuation dimension number set in the collected feature vector to zero to form an edge incomplete feature vector, extract the sector identifier of the edge detection sector where the target to be detected is located, update the AI ​​observation record to include the edge incomplete feature vector, the sector identifier, the AI ​​camera identifier and the observation time, and mark the AI ​​observation record as an edge type.

[0065] S35, the cloud server stores AI observation records labeled as center type and AI observation records labeled as edge type in a real-time observation data cache. Further, the cloud server selects two AI observation records from the real-time observation data cache; the AI ​​observation record with the earlier observation time is designated as first observation information, and the AI ​​observation record with the later observation time is designated as second observation information. It should be noted that the first observation information and the second observation information come from the same AI camera or different AI cameras. In some embodiments, when a narrowband IoT sensor detects a trigger event, it reports a trigger event record including the narrowband IoT sensor identifier and the trigger time to the cloud server. The cloud server stores the received trigger event record in the real-time data cache of the cloud computing platform.

[0066] S36, the cloud server filters AI observation records whose observation times fall within a preset time window from the real-time observation data cache. From the filtered AI observation records, the record with the earlier observation time is selected as a candidate starting point AI observation record, and the record with the later observation time is selected as a candidate ending point AI observation record, forming candidate association pairs. It should be noted that the first observation information corresponds to the candidate starting point AI observation record, the second observation information corresponds to the candidate ending point AI observation record, and the preset time window is 120 seconds.

[0067] S37, determine whether the target to be detected in the first observation information is located in the central region or the edge region.

[0068] S38, in response to the target to be detected being located in the edge region, identify the feature incomplete pattern of the first observation information.

[0069] For example, step S38 includes steps S381 to S382, wherein: S381, in response to the candidate starting point AI observation record being of the edge type, extract the edge incomplete feature vector from the candidate starting point AI observation record, and identify the feature dimension number with a value of zero in the edge incomplete feature vector to form a set of starting point missing dimension numbers.

[0070] S382, extract sector identifiers from the candidate starting point AI observation records, and concatenate the sector identifiers with the starting point missing dimension number set to form a starting point edge missing identifier.

[0071] S39, query the auxiliary detection information of the incomplete feature pattern mapping according to the first mapping relationship, and generate the compensated observation information based on the auxiliary detection information.

[0072] For example, step S39 includes steps S391 to S395, wherein: S391, query the starting edge missing identifier in the edge sequence mapping table to obtain the expected sensor identifier sequence mapped by the starting edge missing identifier.

[0073] S392, Obtain the actual auxiliary detection information between the first observation information and the second observation information.

[0074] For example, step S392 includes steps S3921 to S3922, wherein: S3921, extract the observation time from the candidate starting point AI observation record and record it as the starting point observation time, and extract the observation time from the candidate ending point AI observation record and record it as the ending point observation time.

[0075] S3922, Filter trigger event records whose trigger time is between the starting observation time and the ending observation time from the real-time data cache of the cloud server, and extract the narrowband IoT sensor identifiers from the trigger event records to form a measured sensor identifier sequence.

[0076] S393, Based on the second mapping relationship and combined with the measured sensor identifier sequence, query the feature differences mapped by the measured sensor identifier sequence.

[0077] In some embodiments, the measured sensor identifier sequence is queried in the sequence compensation mapping table. Further, when a mapping relationship exists in the sequence compensation mapping table that perfectly matches the measured sensor identifier sequence, the feature dimension difference vector in that mapping relationship is extracted. Even further, when no perfectly matching mapping relationship exists in the sequence compensation mapping table, the length of the longest common subsequence between the measured sensor identifier sequence and each historical sensor identifier sequence in the sequence compensation mapping table is calculated, and the historical sensor identifier sequence with the largest longest common subsequence length is selected, and the feature dimension difference vector corresponding to that historical sensor identifier sequence is extracted. It should be noted that narrowband IoT sensors may experience missed detections or false detections, leading to differences between the measured sensor identifier sequence and historical sensor identifier sequences. The longest common subsequence retains the longest identical portion appearing in relative order in the two sequences, filtering out the historical records with the most similar spatial location sequences.

[0078] In other embodiments, step S393, which queries feature differences based on the second mapping relationship through trajectory shape encoding matching, includes steps S3931 to S3936, wherein: S3931, Query the deployment coordinates corresponding to each narrowband IoT sensor identifier in the measured sensor identifier sequence from the sensor deployment database, and sort them by trigger time to form a measured sensor coordinate sequence. It should be noted that the measured sensor coordinate sequence consists of M deployment coordinates sorted by trigger time.

[0079] S3932, calculate the spatial displacement vector between adjacent deployment coordinates in the measured sensor coordinate sequence, perform direction normalization processing on each spatial displacement vector, and concatenate the normalized spatial displacement vectors in the order of trigger time to form a measured trajectory shape encoding vector. In some embodiments, the i-th deployment coordinate and the (i+1)-th deployment coordinate in the measured sensor coordinate sequence constitute a set of adjacent deployment coordinates, and the spatial displacement vector between them represents the spatial orientation change of the target to be detected from the i-th sensor position to the (i+1)-th sensor position. M deployment coordinates generate M-1 spatial displacement vectors. Further, the direction normalization processing extracts the direction information of the spatial displacement vector and eliminates the influence of the absolute distance between adjacent sensors in the spatial displacement vector, so that the normalized spatial displacement vector only represents the target's motion direction. It should be noted that the direction normalization process frees the measured trajectory shape encoding vector from the constraint of sensor deployment spacing. In the monitoring area, when two target motion paths have the same direction change pattern but different deployment spacing between the sensors they pass through, the spatial displacement vector sequences before normalization are different, while the spatial displacement vector sequences after normalization tend to be consistent, so that motion paths with the same direction change pattern can be classified into the same trajectory shape category.

[0080] S3933, extract the historical sensor identifier sequence from each mapping relationship in the sequence compensation mapping table, and generate the historical trajectory shape encoding vector for each mapping relationship according to the methods of steps S3931 and S3932. In some embodiments, when the cloud server establishes the sequence compensation mapping table in step S2, it simultaneously performs the processing flow of steps S3931 and S3932 on the historical sensor identifier sequence in each mapping relationship, and stores the generated historical trajectory shape encoding vector as an additional field of the mapping relationship in the sequence compensation mapping table.

[0081] S3934, the measured trajectory shape encoding vector is matched with the historical trajectory shape encoding vector of each mapping relationship in the sequence compensation mapping table. In some embodiments, when the measured trajectory shape encoding vector and the historical trajectory shape encoding vector have the same dimension, the spatial distance between the two vectors is calculated using Euclidean distance metric. Further, when the measured trajectory shape encoding vector and the historical trajectory shape encoding vector have different dimensions, a dynamic time warping algorithm is used to calculate the spatial distance between the two vectors. It should be noted that the length of the measured sensor identification sequence and the length of the historical sensor identification sequence may differ due to missed detections by narrowband IoT sensors, resulting in inconsistent dimensions of the corresponding trajectory shape encoding vectors. The dynamic time warping algorithm flexibly aligns two spatial displacement vector sequences of unequal lengths, finding the alignment path with the minimum cumulative spatial distance under the condition of allowing local temporal scaling, and is suitable for scenarios where inconsistent sequence lengths are triggered by missed sensor detections.

[0082] S3935, select the mapping relationship corresponding to the historical trajectory shape encoding vector with the smallest spatial distance, and extract the feature dimension difference vector from the mapping relationship. In some embodiments, when there are multiple mapping relationships where the historical trajectory shape encoding vectors have the same spatial distance as the measured trajectory shape encoding vector and are all minimum values, select the mapping relationship with the largest longest common subsequence length between the historical sensor identifier sequence and the measured sensor identifier sequence from these mapping relationships, and extract the feature dimension difference vector corresponding to the mapping relationship.

[0083] S3936, the feature dimension difference vector extracted in step S3935 is passed to step S394 to perform feature fusion and generate compensated observation information.

[0084] Through the implementation of steps S3931 to S3936, the matching basis in the sequence compensation mapping table is transformed from discrete comparison of sensor identifiers to geometric shape matching of the target motion trajectory in continuous space. When the identifier of a narrowband IoT sensor in the monitoring area is changed due to equipment maintenance or upgrade, the spatial displacement vector generated by the new sensor deployed in the same spatial location in the trajectory shape encoding is consistent with that of the original sensor. The mapping relationship established based on trajectory shape encoding in the sequence compensation mapping table does not need to be rebuilt due to the change of sensor identifier, avoiding the interruption of edge area feature compensation function due to the inability to match historical compensation data with the new sensor identifier during the sensor replacement transition period. In addition, when multiple targets to be detected move along trajectories with different sensor identifier paths but similar movement direction change patterns, trajectory shape encoding matching can identify the consistency of these trajectories at the spatial geometric level and associate them with the same feature dimension difference vector for compensation, expanding the effective applicability of each mapping relationship in the sequence compensation mapping table.

[0085] S394, the first observation information and the feature difference are fused to generate compensated observation information. For example, an edge-damaged feature vector is extracted from the first observation information, and the edge-damaged feature vector is superimposed with the feature dimension difference vector. The non-zero feature dimensions in the edge-damaged feature vector retain their original values, while the zero-valued feature dimensions are filled with the values ​​of the same feature dimension from the feature dimension difference vector, thus generating a compensated fused feature vector.

[0086] S395, the compensated fusion feature vector, the AI ​​camera identifier of the first observation information, and the observation time are combined to form compensated observation information.

[0087] By implementing step S3, the sensor trigger sequence is used to compensate for the missing visual feature dimensions of the camera edge area, so that the edge area observation, which was originally unable to be associated due to feature incompleteness, can also participate in target tracking, thereby expanding the effective monitoring range of the camera and making the observation information of the edge area of ​​the field of view usable, thus avoiding the tracking loss of the target to be detected due to the failure of edge observation when the camera switches.

[0088] S4: In response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition, a target trajectory record is generated, and the feature compensation association data is updated based on the target trajectory record; the first observation information and the second observation information are acquired through a visual acquisition device, and the auxiliary detection information is acquired through an Internet of Things sensor.

[0089] In some embodiments, the same target condition is determined based on feature similarity. For example, step S4 includes steps S40 to S48, wherein: S40, determine the preset similarity threshold and the feature similarity calculation method. In some embodiments, feature similarity is calculated using a cosine similarity algorithm. The preset similarity threshold is determined based on ROC curve analysis of historical monitoring big data by a cloud computing platform, and the similarity corresponding to the point with the largest difference between the true positive rate and the false positive rate on the ROC curve is selected as the preset similarity threshold. Optionally, the preset similarity threshold is between 0.75 and 0.85, and preferably 0.80 in monitoring scenarios combining AI and IoT.

[0090] S41, when both the candidate starting point AI observation record and the candidate ending point AI observation record are of the center type, extract the center region feature vector from the candidate starting point AI observation record and extract the center region feature vector from the candidate ending point AI observation record. Calculate the similarity between the two center region feature vectors using a cosine similarity algorithm. When the similarity is greater than the preset similarity threshold, determine that the first observation information and the second observation information satisfy the same target condition.

[0091] S42, when the candidate starting point AI observation record is of edge type and the candidate ending point AI observation record is of center type, extract the center region feature vector from the candidate ending point AI observation record, calculate the similarity between the compensation fusion feature vector generated in step S394 and the center region feature vector using the cosine similarity algorithm, and when the similarity is greater than the preset similarity threshold, determine that the compensation observation information and the second observation information satisfy the same target condition.

[0092] S43, when the candidate starting point AI observation record is of the center type and the candidate ending point AI observation record is of the edge type, feature compensation is performed on the second observation information to generate second compensated observation information, and it is determined whether the first observation information and the second compensated observation information satisfy the same target condition. It should be noted that the process of feature compensation for the second observation information is the same as the process of feature compensation for the first observation information in step S3.

[0093] S44, when both the candidate starting point AI observation record and the candidate ending point AI observation record are edge types, the compensation fusion feature vector generated in the identification step S394 and the feature dimension numbers in the compensation fusion feature vector in the second compensation observation information, which are all non-zero values, form a common effective feature dimension number set. The feature dimension values ​​of the two compensation fusion feature vectors in the common effective feature dimension number set are extracted, and the similarity is calculated by the cosine similarity algorithm to determine whether they meet the same target condition.

[0094] S45, for candidate association pairs that meet the same target conditions, extract the AI ​​camera identifier and the starting observation time from the candidate starting point AI observation record, extract the AI ​​camera identifier and the ending observation time from the candidate ending point AI observation record, generate a target movement trajectory record including the AI ​​camera identifier, the starting observation time, the ending observation time and the measured sensor identifier sequence, and output the target movement trajectory record to the monitoring terminal.

[0095] S46, add the candidate starting point AI observation record, the candidate ending point AI observation record, and the measured sensor identification sequence to the historical monitoring data.

[0096] S47, Update the feature compensation association data based on the updated historical monitoring data. For example, the cloud server invokes cloud computing resources to execute the processing flow of step S2 on the updated historical monitoring data, re-establishing the sequence compensation mapping table and the edge sequence mapping table.

[0097] In some embodiments, step S47 is followed by cross-camera collaborative compensation update steps S471 to S476, wherein: S471: Extract the type markers of the candidate starting point AI observation record and the candidate ending point AI observation record from the target movement trajectory record generated in step S45, and determine the type marker combination. Further, in response to the candidate starting point AI observation record being of edge type and the candidate ending point AI observation record being of center type, execute steps S472 to S476. Even further, in response to the candidate starting point AI observation record being of center type and the candidate ending point AI observation record being of edge type, swap the roles of the candidate starting point AI observation record and the candidate ending point AI observation record in steps S472 to S476 before execution. It should be noted that the execution of step S471 is premised on the fact that in step S4 it has been determined that the candidate association pair satisfies the same target condition and that the target movement trajectory record has been generated in step S45.

[0098] S472, extract the central region feature vector as a collaborative reference feature vector from the AI ​​observation record tagged as center type, and extract the compensation fusion feature vector from the compensation observation information generated in step S395. It should be noted that the collaborative reference feature vector is the complete visual feature vector extracted when the same target to be detected is observed in the central region of the AI ​​camera, and all feature dimensions are valid values.

[0099] S473, identify the feature dimension numbers filled by the feature dimension difference vector in step S394 in the compensated fusion feature vector to form a compensation dimension number set, and identify the feature dimension numbers in the compensated fusion feature vector that retain the original non-zero values ​​of the edge-damaged feature vector to form a retained dimension number set. It should be noted that the compensation dimension number set is consistent with the attenuation dimension number set determined in step S344. The feature dimension values ​​in the retained dimension number set are the original acquired values ​​of the edge-damaged feature vector in the first observation information, without undergoing the compensation processing of step S394.

[0100] S474: Extract the feature dimension values ​​corresponding to the retained dimension number set from the compensated fusion feature vector to form a compensated retained sub-vector; extract the feature dimension values ​​corresponding to the retained dimension number set from the collaborative reference feature vector to form a reference retained sub-vector; calculate the similarity between the compensated retained sub-vector and the reference retained sub-vector using a cosine similarity algorithm, as a measure of retained dimension consistency. It should be noted that the feature dimensions within the retained dimension number set originate from the original acquired values ​​in the edge-damaged feature vector that were not zeroed in step S345, and participate in the determination of the same target condition as a component of the visual feature vector in the compensation matching process of steps S3 to S4. In step S474, the feature dimensions within the retained dimension number set are transformed from participating elements in the determination of the same target into verification criteria for collaborative compensation triggering conditions. The original reliable feature dimensions from edge region observations, without compensation processing, are used to independently verify whether two AI observation records truly belong to the same target outside the compensation process.

[0101] S475, in response to the retained dimension consistency measure satisfying a preset collaborative compensation condition, the feature dimension values ​​corresponding to the compensation dimension number set are extracted from the collaborative reference feature vector, and the feature dimension values ​​corresponding to the compensation dimension number set in the compensation fusion feature vector are replaced to generate a collaborative compensation feature vector. In some embodiments, the preset collaborative compensation condition is that the retained dimension consistency measure reaches a preset collaborative compensation threshold, and the value of the preset collaborative compensation threshold is higher than the preset similarity threshold determined in step S40. It should be noted that the preset collaborative compensation threshold is higher than the preset similarity threshold, and a more stringent consistency verification is applied to the retained dimensions based on the same target determination in step S4, reducing the risk of writing incorrect feature values ​​in collaborative compensation. Further, the feature dimension values ​​corresponding to the retained dimension number set in the collaborative compensation feature vector retain the original acquired values ​​in the compensation fusion feature vector, and the feature dimension values ​​corresponding to the compensation dimension number set are replaced with the true feature values ​​of the same target to be detected when observed in the central region in the collaborative reference feature vector. Furthermore, in response to the fact that the retention dimension consistency metric does not meet the preset collaborative compensation condition, collaborative compensation is not performed, and the compensation fusion feature vector generated in step S394 is retained as the final compensation result of the candidate association pair, and step S48 is executed directly.

[0102] S476, Update the mapping relationship matched in step S393 in the sequence compensation mapping table based on the collaborative compensation feature vector.

[0103] For example, step S476 includes steps S4761 to S4763, wherein: S4761, following the method for generating the feature dimension difference vector in step S222, the collaborative reference feature vector is used as historical complete feature information, and the edge-damaged feature vector in the first observation information is used as historical incomplete feature information to generate a collaborative correction difference vector. It should be noted that the data source for the collaborative correction difference vector in step S4761 is the actual feature records observed in the central and edge regions of the same target during the current monitoring process. The feature dimension values ​​corresponding to the compensation dimension number set in the collaborative correction difference vector reflect the real feature changes of the specific target from the central region to the edge region, rather than the statistical feature changes of a historical target group.

[0104] S4762, locate the mapping relationship matched in step S393 in the sequence compensation mapping table, and store the collaborative correction difference vector as the correction difference vector field of the mapping relationship. It should be noted that step S4762 adds a correction difference vector field to the mapping relationship. This field is stored independently of the original feature dimension difference vector and does not cover the original feature dimension difference vector.

[0105] S4763, mark the mapping relationship as corrected. When a new target to be detected finds the mapping relationship in step S393, in response to the mapping relationship being in a corrected state, extract the corrected difference vector to replace the original feature dimension difference vector and pass it to step S394 to perform feature fusion.

[0106] In some embodiments, when the same mapping relationship is triggered multiple times in steps S471 to S476 and the preset collaborative compensation condition is met each time, the collaborative correction difference vector generated later replaces the correction difference vector stored previously. Further, when the cumulative number of times the same mapping relationship is triggered in steps S471 to S476 reaches a preset correction count threshold, the correction difference vector is overwritten into the original feature dimension difference vector field, the correction difference vector field is deleted, and the mapping relationship is marked as permanently corrected. It should be noted that the preset correction count threshold is set to distinguish between individual feature differences from a single observation and stable feature differences verified by multiple observations. By storing the correction difference vector field independently rather than directly overwriting the original feature dimension difference vector, the original statistical difference vector is retained as a fallback option. When the cumulative correction count reaches the preset correction count threshold, the collaborative correction results of multiple different targets mutually confirm the stability of the feature degradation mode of the mapping relationship in the edge region. At this time, the correction difference vector is permanently added to the original field. Optionally, the preset correction count threshold is 3 to 5, preferably 3 in monitoring scenarios combining AI and IoT.

[0107] Through the implementation of steps S471 to S476, the feature dimension difference vector stored in the sequence compensation mapping table is gradually replaced by the real feature difference values ​​of the same target in the central region and the edge region from the historical target group statistical estimate value established at the beginning of step S2. This improves the filling accuracy of the missing dimension in the compensation fusion feature vector from the group statistical level to the individual observation level, and reduces the misjudgment and missed judgment of the same target caused by the statistical bias of the compensation features in step S4.

[0108] Through the implementation of step S4, the successfully associated target trajectory information is fed back to the feature compensation association data, so that the sequence compensation mapping table and the edge sequence mapping table continuously accumulate new mapping relationships as the monitoring system continues to run, forming a closed-loop optimization mechanism from feature compensation to target association and then to data update.

[0109] In some embodiments, step S47 is followed by step S48, wherein: S48, based on the updated feature compensation association data, predict the sensor location that the target to be detected may trigger next.

[0110] For example, step S48 includes steps S481 to S485, wherein: S481, Extract endpoint observation information from the target movement trajectory record. In some embodiments, the cloud server extracts the endpoint observation time and the AI ​​camera identifier of the candidate endpoint AI observation record from the target movement trajectory record.

[0111] S482, query multiple historical auxiliary detection information in the updated feature compensation association data. In some embodiments, query all mapping relationships in the updated edge sequence mapping table.

[0112] S483, extract the first detection unit identifier from each historical auxiliary detection information. In some embodiments, extract the first narrowband IoT sensor identifier from each queried mapping relationship.

[0113] S484, count the frequency of occurrence of each of the first detection unit identifiers, and select the first detection unit identifier with the highest frequency as the predicted detection unit identifier. Further, when multiple first detection unit identifiers have the same frequency and are all the highest frequency, select the first detection unit identifier closest to the AI ​​camera position of the endpoint observation information as the predicted detection unit identifier. Even further, query the deployment coordinates of the predicted detection unit identifier in the sensor deployment database.

[0114] S485, Generate and output a prediction record including the prediction detection unit identifier, the deployment coordinates, and the endpoint observation information. In some embodiments, generate a target prediction position record including the prediction detection unit identifier, the deployment coordinates, and the endpoint observation time, and output the target prediction position record to a monitoring terminal for target prediction position display.

[0115] Through the implementation of steps S481 to S485, the sensor location that may be triggered by the next step of the target to be detected is predicted based on the updated feature compensation association data, so that the monitoring system changes from passive response to active early warning, and gains a time window for security personnel to handle the situation on-site or for the linkage operation of the automatic control system to deploy control in advance.

[0116] Example 2 is an embodiment of the present invention, which provides a combined application system of cloud computing and big data based on the combination of AI and IoT, including: Visual acquisition devices, IoT sensors, and cloud servers.

[0117] In some embodiments, the visual acquisition device is an AI camera, and the IoT sensor is a narrowband IoT sensor.

[0118] Furthermore, the observation area is the field of view of the AI ​​camera.

[0119] It should be noted that the AI ​​camera is connected to a cloud server via a network, and the captured image frames are uploaded to the cloud server in real time. The narrowband IoT sensor is connected to the cloud server via an NB-IoT network and reports the trigger event record when a trigger event is detected.

[0120] The cloud server includes: The region division module is used to divide the observation area into a central region and an edge region, wherein the observation area is the field of view of the AI ​​camera for the target monitoring scene; The association establishment module is used to establish feature compensation association data, which includes a first mapping relationship. The first mapping relationship is used to map the feature incomplete pattern of the edge region and auxiliary detection information. The feature compensation module is used to respond to the fact that the target to be detected is located in the edge region, query the feature incomplete pattern of the first observation information of the target to be detected according to the first mapping relationship, obtain the auxiliary detection information mapped by the feature incomplete pattern, and generate compensated observation information based on the auxiliary detection information. The determination module is used to generate a target trajectory record in response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition; The update module is used to update the feature compensation association data based on the target trajectory record.

[0121] In some embodiments, the cloud server further includes: The big data storage module is used to store historical monitoring data, sequence compensation mapping tables, and edge sequence mapping tables; The real-time data caching module is used to cache AI ​​observation records and triggered event records; A sensor deployment database is used to store the identifiers, deployment coordinates, and detection range parameters of narrowband IoT sensors.

[0122] The region segmentation module reads historical monitoring data from the big data storage module, determines the preset distance threshold and edge detection sector segmentation parameters, and provides the segmentation results to the feature compensation module.

[0123] The association establishment module reads historical monitoring data from the big data storage module, establishes the sequence compensation mapping table and the edge sequence mapping table, and stores them in the big data storage module for querying by the feature compensation module.

[0124] The feature compensation module obtains AI observation records and trigger event records from the real-time data cache module, queries the sequence compensation mapping table and the edge sequence mapping table from the big data storage module, queries the expected trigger sensor set from the sensor deployment database according to the location of the target to be detected, and generates compensation observation information to be provided to the judgment module.

[0125] The update module receives candidate association pairs that meet the same target condition output by the determination module, generates target movement trajectory records and stores them in the big data storage module, and triggers the association establishment module to update the sequence compensation mapping table and the edge sequence mapping table.

[0126] This embodiment also provides an electronic device applicable to a method for combining cloud computing and big data in an AI and IoT scenario, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for combining cloud computing and big data in an AI and IoT scenario as proposed in the above embodiment.

[0127] This embodiment also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements a method for combining cloud computing and big data in a scenario combining AI and IoT, as proposed in the above embodiment.

[0128] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for combining cloud computing and big data in the scenario of AI and IoT proposed in the above embodiment. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0129] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for combining cloud computing and big data in a scenario combining AI and IoT, characterized in that, include: The observation area is divided into a central area and an edge area, which is the field of view of the AI ​​camera for the target monitoring scene; Establish feature compensation association data, which includes a first mapping relationship, the first mapping relationship being used to map the feature incompleteness pattern of the edge region and auxiliary detection information; In response to the target being detected being located in an edge region, compensated observation information of the target being detected is obtained. The process of obtaining the compensated observation information of the target being detected includes: querying the feature incomplete pattern of the first observation information of the target being detected according to the first mapping relationship, obtaining auxiliary detection information mapped by the feature incomplete pattern, and generating the compensated observation information based on the auxiliary detection information. In response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition, a target trajectory record is generated, and the feature compensation association data is updated based on the target trajectory record; The first observation information and the second observation information are acquired through a visual acquisition device, and the auxiliary detection information is acquired through an Internet of Things (IoT) sensor.

2. The method for combining cloud computing and big data in a scenario combining AI and IoT as described in claim 1, characterized in that, The process of generating the compensated observation information based on the auxiliary detection information includes: Obtain the actual auxiliary detection information between the first observation information and the second observation information; Based on the second mapping relationship, and combined with the actual auxiliary detection information, query the feature differences mapped by the actual detection information; The first observation information and the feature difference are fused to generate the compensated observation information. The effective features of the first observation information in the compensated observation information retain their original values, and the missing features of the first observation information are filled by the feature difference.

3. The method for combining cloud computing and big data in a scenario combining AI and IoT, as described in claim 2, is characterized in that... The division of the edge region also includes: The edge region is divided into multiple edge sub-regions along the azimuth angle based on the observation parameters; The incomplete feature pattern includes edge sub-region identifiers and feature missing information.

4. The method for combining cloud computing and big data in a scenario combining AI and IoT as described in claim 3, characterized in that, The determination of the same target condition includes: When both the first observation information and the second observation information are complete feature information, determine whether the first observation information and the second observation information satisfy the same target condition; When the first observation information is incomplete feature information and the second observation information is complete feature information, determine whether the compensation observation information and the second observation information satisfy the same target condition. When the first observation information is complete feature information and the second observation information is incomplete feature information, feature compensation is performed on the second observation information to generate second compensated observation information, and it is determined whether the first observation information and the second compensated observation information satisfy the same target condition. When both the first observation information and the second observation information are incomplete feature information, it is determined whether the common valid features of the compensated observation information and the second compensated observation information satisfy the same target condition.

5. The method for combining cloud computing and big data in a scenario combining AI and IoT as described in claim 4, characterized in that, After updating the feature compensation association data based on the target trajectory record, the method further includes: Extract endpoint observation information from the target trajectory record; Query multiple historical auxiliary detection information in the updated feature compensation association data; Extract the first detection unit identifier from each of the historical auxiliary detection information; The frequency of occurrence of each of the first detection unit identifiers is counted, and the first detection unit identifier with the highest frequency of occurrence is selected as the predicted detection unit identifier; Generate and output a prediction record that includes the prediction detection unit identifier and the endpoint observation information. Generate and output a prediction record that includes the prediction detection unit identifier and the endpoint observation information.

6. The method for combining cloud computing and big data in a scenario combining AI and IoT as described in claim 5, characterized in that, The establishment of feature-compensated associated data includes: Extract complete historical feature information, incomplete historical feature information, and historical auxiliary detection information from historical data; Calculate the feature difference between the complete historical feature information and the incomplete historical feature information, and establish a second mapping relationship between the historical auxiliary detection information and the feature difference; Extract the incomplete feature patterns from the historical incomplete feature information, and establish the first mapping relationship between the incomplete feature patterns and the historical auxiliary detection information; The feature compensation association data includes the first mapping relationship and the second mapping relationship.

7. The method for combining cloud computing and big data in a scenario combining AI and IoT as described in claim 6, characterized in that, The first observation information is the visual feature information of the target to be detected collected by the AI ​​camera at the first observation time; the second observation information is the visual feature information of the target to be detected collected by the AI ​​camera at the second observation time. The auxiliary detection information is the detection information reported by the IoT sensor between the first observation time and the second observation time; The observation parameters are the field-of-view parameters of the visual acquisition device.

8. A cloud computing and big data integrated application system based on AI and IoT scenarios, using the method described in any one of claims 1 to 7, characterized in that, include: Visual acquisition devices, IoT sensors, and cloud servers; The cloud server includes: The region division module is used to divide the observation area into a central region and an edge region, wherein the observation area is the field of view of the AI ​​camera for the target monitoring scene; The association establishment module is used to establish feature compensation association data, which includes a first mapping relationship. The first mapping relationship is used to map the feature incomplete pattern of the edge region and auxiliary detection information. The feature compensation module is used to respond to the fact that the target to be detected is located in the edge region, query the feature incomplete pattern of the first observation information of the target to be detected according to the first mapping relationship, obtain the auxiliary detection information mapped by the feature incomplete pattern, and generate compensated observation information based on the auxiliary detection information. The determination module is used to generate a target trajectory record in response to the compensation observation information and the second observation information of the target to be detected satisfying the same target condition; The update module is used to update the feature compensation association data based on the target trajectory record.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for combining cloud computing and big data in a scenario combining AI and IoT, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for combining cloud computing and big data in a scenario combining AI and IoT, as described in any one of claims 1 to 7.

Images